2012 International Conference on Industrial and Intelligent Information (ICIII 2012)
IPCSIT vol. XX (2011) © (2011) IACSIT Press, Singapore
Three-Dimensional Morphology Quantification of Biofilm Structures
from Confocal Laser Scanning Microscopy Images
Jerzy S. Zielinski1, Agnieszka K. Zielinska2, Nidhal Bouaynaya1
1
2
University of Arkansas in Little Rock
University of Arkansas for Medical Sciences
Abstract. Staphylococcus aureus is an opportunistic human pathogen and a causative agent of a wide range
of diseases. Its hallmark feature is a capability to form biological structures called biofilms, in which bacteria
can embed themselves in a protective matrix. The study of biofilm structures and morphologies is central to
understand the biological characteristics of the biofilm including its resistance to drugs. These studies,
however, have focused on two-dimensional (2D) analysis of the biofilm images. In this paper, we show the
importance of three-dimensional analysis of the biofilm morphology by taking into account the temporal
information between the images. Specifically, we show that the morphological features computed from a
simple gradient-based three-dimensional segmentation method outperforms not only its 2D homologue but
also sophisticated two-dimensional segmentation methods like the watershed algorithm. Our analysis is
applied to real biofilm confocal laser scanning microsopy images with a ground truth delineated manually by
expert biologists at the University of Arkansas for Medical Sciences.
Keywords: Staphylococcus aureus, Confocal Laser Scanning Microscopy, three-dimensional
segmentation.
1. Introduction
Staphylococcus aureus is an opportunistic human pathogen responsible for diseases that vary in clinical
presentation and severity, ranging from minor skin infections to life-threatening conditions such as
pneumonia, endocarditis and toxic shock syndrome. Increasing rates of staphylococcal infections
combined with dramatic increase of number of infections caused by multi-antibiotic resistant strains led
to the situation in which MRSA (methicillin-resistant S. aureus) mortality rates surpassed those of HIV
in the USA (Nygaard rt al., 2008). For that reason, there is a strong need to understand pathogenesis of
Staphylococcus aureus infections. Studies have shown that there are two major elements contributing to
the fact that these life-threatening infections are so difficult to treat. The first is the emergence of new
antibiotic-resistant strains and the second is that many S. aureus infections are associated with the
formation of a biofilm, which limits the efficacy of any antimicrobial therapy. Biofilms shield
microorganisms from stressful environmental conditions including the defensive mechanisms of the host
immune system and antimicrobials. Since biofilm formation is such a detrimental element of S. aureus
infection, it requires extensive studies. One of the most important parameters of biofilm that is used in
biological studies is the area occupied by the biomass, i.e. biomass volume. Due to the fact that the
biofilm is a porous structure, in order to compute biomass volume, it is necessary to differentiate the area
occupied by the biofilm from the surrounding noisy background.
One of the most comprehensive imaging techniques used for the assessment of the biofilm structure is
the confocal laser scanning microscopy (Lawrence et al., 2002). By using several fluorescent stains or
conjugated antibodies in combination with multichannel CLSM 3D, the location of different biofilm
constituents can be recorded. The majority of confocal laser scanning microscopy (CLSM) image
quantification methodologies require regionsegmentation, i.e. the identification of similar regions and the
differentiation of the signal of interest in the image from the background. There are several methods
described in the literature, which target the problem of biological structure recognition from CLSM
images. Most of these methods, however, rely on a two dimensional segmentation of the images in order
to extract morphological features of the biofilm (Zielinski at al., 2011). Sima et al. presented an
automatic segmentation of small 3D objects using multiscale wavelet edge detection method, where
edges are detected by filtering the image with low-pass filters of different sizes followed by an
application of the gradient operator. This method, although 3D, is suitable for branch-like structures like
neurons, but fails in the recognition of the more irregular and compact shapes formed by biofilms. In this
paper, we perform 3D Biofilm Morphology Quantification
The three dimensional morphology of biofilms can be quantified using volumetric parameters, which are
calculated from voxels in the 3D structure of the biofilm. In this paper, we use the average run lengths
(in X, Y and Z direction), the aspect ratio, the average and maximum diffusion distances (Haluk et al.,
2004), biomass and average thickness as important clinical features of the biofilm.
1.1. Average Run Length
The run length measures the number of consecutive cluster pixels in a given direction, and an average
run length measures the average length of consecutive cluster pixels in the measurement direction. For
instance, if a biofilm has a large average X-run length and a small average Y-run length, then the biofilm
appears stretched in the X direction. This may indicate special growth patterns of the biofilm (Haluk et
al., 2004)
1.2. Aspect Ratio
The aspect ratio is defined as the ratio of the average X-run length to the average Y-run length along the
Y-axis. It indicates the symmetry of growth in the X and Y directions. In particular, significant variation
of the aspect ratio may indicate the presence of stress or micro-colonies that affect the normal growth of
the biofilm (Haluk et al.,2004)
1.3. Average and Maximum Diffusion Distance
The diffusion distance is the minimum linear distance from a cluster voxel to the nearest void voxel in a
three-dimensional structure. The average diffusion distance is the average of all diffusion distances. The
maximum diffusion distance is defined as the distance to the most remote pixel in the cell cluster from a
void cluster. (Haluk et al., 2004). Biologically, the diffusion distance indicates the distance over which
the biofilm has to diffuse in the cell structure.
1.4. Biomass
The biomass in bio-volume, V, is measured from numeric integration of the area of microbial
colonization profiles, following a method previously described in (Keuhn et al.,1998)
1.5. Average Thickness
The average thickness of the biofilm is calculated as the average value of the height of all clusters of the
biofilm rise from solid-substratum in the z direction between cross-sections.
Three-Dimensional Segmentation is the accurate segmentation of highly porous objects from a noisy
background, e.g., bacterial cells in a biofilm from the medium, is a challenging task in image processing.
We show that a simple three-dimensional (3D) segmentation, which relies on the 3D gradient,
outperforms the status-quo two-dimensional segmentation techniques used for the delineation of biofilms
(Zielinski et al., 2011). The power of 3D segmentation stems from the fact that it takes into account the
temporal correlation in addition to the spatial correlation accounted for in 2D analysis.
The gradient field finds edges in the image by treating the set of image as a 3D surface or a video. The
3D region recognition and image segmentation follows this simple two-threshold level procedure:
1. Compute the 3D gradient field on volume created by stacking CLSM images as follows:
where F denotes the 3D image surface or the set of 2D images.
2. Threshold the gradient magnitude in order to find the boundaries of 3D structures buried in the noisy
background.
3. Compute the arithmetic mean of intensity of the obtained binary image surface in order to refine the
threshold level.
4. Threshold the volume into binary values based on the calculated level of refined threshold.
2. Application to CLSM Biofilm Images
2.1. Biofilm culture preparation and image acquisition
Costar 3596 plates (Corning Life Sciences, Acton, MA) wells were coated overnight at 4oC with 20%
human plasma (Sigma) in bicarbonate buffer. Overnight cultures of S. aureus grown in biofiolm media
(TSB-NaCl/Glc) were diluted to an OD600 of 0.05 in fresh TSB-NaCl/Glc. After removing human plasma,
200 ul of each culture was transferred to wells and incubated for 24 h at 37oC. The next day, the wells
were gently washed three times with 0.85% (wt/vol) NaCl, followed by staining with mixture of SYTO9 and propidium iodide diluted in 0.85% (wt/vol) NaCl (Cat\# L7012; Invitrogen, Carlsbad, CA) for
18min. After removing the stain, the wells were gently washed once with 0.85% (wt/vol) NaCl. Biofilm
images were collected by CLSM using a LSM 510 META confocal scanning system (Zeiss, Thornwood,
NY) and Axiovert 200 motorized inverted microscope (Zeiss). SYTO 9 (green, live cells) fluorescence
was detected by excitation at 488 nm, and emission was collected with a 500- to 530-nm bandpass filter.
Propidium Iodide (red; dead cells) fluorescence was detected by excitation at 488 nm and emission was
collected with a 565-615 nm bandpass filter. All z-sections were collected at 1- um intervals by using a
C-Apochromat 40x/1.2W H2O objective lens. Image acquisition and processing was performed by using
an LSM Image Browser (Zeiss).
2.2. Segmentation and parameter quantification results
Z-stack Confocal Scanning Microscopy Images of three different S. aureus colonies were processed
using 2D segmentation and 3D segmentation methods separately. The expert biologist, Agnieszka
Zielinska, performed a (tedious) manual segmentation of the biofilm to be used as the ground truth. The
segmentation and parameter quantification results of 3D and 2D methods are assessed using the mean
square error criterion. Figure 1 shows the output of the 2D and 3D segmentations of a stack of CLSM
images. It is clear that the 3D segmentation better captures the biofilm surface than its 2D homologue.
Figures 2 and 3 further shows the mean square error of the 2D gradient-based segmentation, the 2D
watershed segmentation (Zielinski at al., 2011 and Meyer et al., 1994), and the 3D gradient-based
segmentation. For all the images we had in our dataset, the error was significantly smaller in 3D
segmentation than in 2D segmentation. We, therefore, expect that the resulting 3D biofilm
quantification parameters are more accurate than their corresponding 2D values. Indeed, this is
confirmed by the results displayed in Tables 1, 2 and 3 for the biofilm parameters: average run length,
aspect ratio, average and maximum diffusions, biomass and average thickness. Amazingly, the simple
3D gradient-based segmentation not only performs better than the 2D gradient-based segmentation but
also shows a higher performance, in terms of mean square error, than the sophisticated 2D watershed
segmentation.
Figure 2 Mean Square Error of 2D gradient-based
segmentation
(black),
2D
watershed
segmentation (green) and 3D (red) gradientbased segmentation applied to CLAS z-stack
of images.
Ground Truth
Figure 3. Combined Mean Square Error of 2D
gradient-based (black), 2D watershed
(green) and and 3D gradient-based
segmentations.
3D- gradient
segmentation
2D-gradient
segmentation
2D Watershed
segmentation
Average Run Length X [um]
12.98
12.45
9.69
12.00
Average Run Length Y [um]
12.47
11.65
9.28
11.10
Average Run Length Z [um]
5.02
4.98
4.20
4.56
Aspect Ratio
1.04
1.06
1.04
1.08
Average Diffusion Distance [um]
3.25
2.56
1.30
2.52
Maximum Diffusion Distance [um]
15.70
15.69
13.20
14.98
Biomass [um^3/um^2]
5.03
4.87
4.56
4.72
Average Thickness [um]
10.47
9.25
8.98
9.02
Table 1: Results of biofilm parameter quantification for Stack 1 for 3D and 2D segmentations in comparison with the
ground truth.
Ground Truth
3D-gradient
2D-gradient
2D Watershed
segmentation
segmentation
segmentation
Average Run Length X [um]
21.32
20.75
19.69
20.09
Average Run Length Y [um]
20.50
20.67
19.38
20.01
Average Run Length Z [um]
6.02
6.98
6.20
6.50
Aspect Ratio
1.04
1.00
1.01
1
Average Diffusion Distance [um]
4.40
4.66
4.30
4.40
Maximum Diffusion Distance [um]
15.96
15.99
15.64
15.70
Biomass [um^3/um^2]
7.59
6.97
5.71
6.30
Average Thickness [um]
12.53
11.38
10.00
11.00
Table 2: Results of biofilm parameter quantification for stack 2 for 3D and 2D segmentations in comparison with the
ground truth.
Ground Truth
3D-gradient
2D-gradient
2D Watershed
segmentation
segmentation
segmentation
Average Run Length X [um]
52.98
50.48
48.00
49.86
Average Run Length Y [um]
50.47
48.93
47.14
48.21
Average Run Length Z [um]
14.02
14.06
13.04
13.85
Aspect Ratio
1.05
1.03
1.02
1.03
Average Diffusion Distance [um]
10.25
9.37
9.10
9.20
Maximum Diffusion Distance [um]
55.70
15.69
13.20
14.97
Biomass [um^3/um^2]
20.65
19.65
18.91
19.03
Average Thickness [um]
24.56
24.00
23.46
23.92
Table 3: Results of biofilm parameter quantification for stack 3 for 3D and 2D segmentations in comparison with the
ground truth.
Figure 1. 2D and 3D-gradient based segmentation: Column 1: the original CLSM images. Column 2: 3D gradient-based
segmentation; Column 3: 2D gradient-based segmentation; Column 4: 2D Watershed Segmentation
3. Conclusion
In this paper, we showed the importance of 3D analysis of biofilm structures, which yields more accurate
morphological parameter quantification for clinical and biological assessment of the biofilm than
sophisticated 2D-based analysis like the watershed segmentation. Two-dimensional analysis of the
biofilm morphology treats the CLSM images independently from each other, whereas the 3D analysis
takes into account the temporal correlation between the stacked images. Even though the proposed 3D
segmentation relies on a simple 3D gradient computation, we showed that the resulting biofilm
parameters are more accurate than sophisticated 2D-based segmentations, like the watershed
segmentation. Accurate computation of biofilm morphological parameters, like biomass, thickness and
diffusion coefficients, are pivotal to understanding biofilm growth and biological characteristics; and
thus design efficient therapeutics to target it.
4. Bibliography and Literature
•
•
Boothby, William; An introduction to differentiable manifolds and Riemannian geometry. Pure and Applied
Mathematics, volume 120 (second ed.). Orlando, FL: Academic Press (1986). ISBN 0-12-116053-X
Haluk Beyenal, Conrad Donovan, Zbigniew Lewandowski, Gary Harkin; Three-dimensional biofilm structure
quantification Journal of Microbiological Methods 59 (2004) 395– 413
•
Heydorn A; Nielsen AT; Hentzer M; Sternberg C; Givskov M; Ersbll BK; Molin S., “Quantification of biofilm
structures by the novel computer program comstat.,” Microbiology, vol. 146, pp. 2395–2407, 2000.
•
Hubbard, J. H.; Hubbard, B. B.. Vector calculus, linear algebra, and differential forms. A unified approach.
Upper Saddle River, NJ (1999): Prentice Hall. ISBN 0-13-657446-7
Lawrence JR; Neu TR; Marshall KC, Colonization, adhesion, aggregation and biofilms, American Society for
Microbiology, 2002.
•
•
Kuehn M; Hausner M; Bungartz HJ, Wagner M; Wilderer PA; Wuertz S, “Automated confocal laser scanning
microscopy and semiautomated image processing for analysis of biofilms.,” Applied and Environmental
Microbiology, vol. 64, no. 11, pp. 4115–4127, 1998.
•
Meyer; Fernand, “Topographic distance and watershed lines.,” Signal Processing, vol. 38, pp. 113–125, 1994.
•
Nygaard TK; DeLeo FR; Voyich JM, “Community-associated methicillin resistant staphylococcus aureus skin
infections: advances toward identifying the key virulence factors.,” Curr Opin Infect Dis, vol. 21, no. 2, pp.
147–152, 2008.
•
Rice KC; Mann EE; Endres JL; Weiss EC; Cassat JE; Smeltzer MS; Bayles KW, “The cida murein hydrolase
regulator contributes to dna release and biofilm development in staphylococcus aureus.,” Proceedings of the
National Academy of Sciences of the United States of America, vol. 104, no. 19, pp. 8112–8118, 2007.
•
Sharpe, R. Differential geometry. Springer-Verlag (1997). ISBN 0-387-94732-9
•
Staudt C; Horn H; Hempel DC; Neu TR., “Volumetric measurements of bacterial cells and extracellular
polymeric substance glycoconjugates in biofilms.,” Biotechnol Bioeng., vol. 88, no. 5, pp. 585–592, 2004.
•
Xavier J.B.; White D.C. ; Almeida J.C., “Automated biofilm morphology quantification from confocal laser
scanning microscopy imaging.,” Water Science and Technology, vol. 47, no. 5, pp. 31–37, 2003.
•
Zielinski J.S., Zielinska A.K., Bouaynaya N, Smeltzer M.S. "Automated Biofilm Region Recognition And
Morphology Quantification From Confocal Laser Scanning Microscopy Imaging", in Biomedical Science and
Engineering Conference - Image Informatics and Analytics in Biomedicine (BSEC 2011), Knoxville,
Tennessee, March 2011.